Physical Artifacts for Agents in a Cyber-PhysicalSystem: A Case Study in Oil & Gas Scenario

(EEAS)1st Fabian Cesar Pereira Brandao Manoel

Federal Center for Technological Education (CEFET/RJ)Rio de Janeiro, Brasil0000-0003-0614-0592

3rd Leandro Marques SamynFederal Center for Technological Education (CEFET/RJ)

Rio de Janeiro, Brasil0000-0002-0733-4172

2nd Carlos Eduardo PantojaFederal Center for Technological Education (CEFET/RJ)

Rio de Janeiro, Brasil0000-0002-7099-4974

4rd Vinicius Souza de JesusFederal Center for Technological Education (CEFET/RJ)

Rio de Janeiro, Brasilsouza.vdj@gmail.com

Abstract—Physical devices have been integrated with artificialintelligence to create Cyber-Physical Systems (CPS). Multi-AgentSystems (MAS) can provide pro-activity and autonomy usingagents, social organizations, and environment modeling by meansof artifacts. Usually, some works that use MAS for interfacingphysical environments employ agents accessing directly all theavailable data of the environment, which could overload thisagent. This issue could be avoided if there were tools to facilitatethe integration of sensors and actuators as artifacts into thephysical environment. Therefore, the objective of this work is tocreate physical artifacts capable of accessing hardware devicesfrom a physical environment to be used by agents in a MAS. Asthe Oil & Gas industry demands robustness in its equipment andan ability to do predictive maintenance, a case study includingMAS and CPS was developed and some tests were carried outto validate the functioning of physical artifacts.

Index Terms—Physical Artifact, Physical environment, Oil &Gas Industry

I. INTRODUCTION

In the last years, the agent approach has been switchingfrom simulated to physical applications where Multi-AgentSystems (MAS) have been used to interact and control devicesworking in dynamic environments [1] [2] [3] [4]. In general,some approaches define four dimensions that guide a MASimplementation: agency, environmental, organisational [5], andinteraction [6]. Agents interact in an environment accordingto their implemented beliefs, desire, and intentions (BDI);Artifacts provide operational functions and observable prop-erties for agents, and they represent non-cognitive entitiessituated in workspaces; organizational dimension models thesociety notion and the collective norms of the agent’s behavior;interaction dimension models the interaction between the threedimensions (agent, environment, and organization). In parallel,when connecting computing elements to physical elements,such as embedded computers connected in a network, it main-tains a system known as Cyber-Physical Systems (CPS) [7].

When considering physical environments, rarely they areexplored considering other dimensions aside from the agentone. In an agent application in the oil domain, only the agencydimension is considered [8]. The agent performance dependsdirectly on the amount of information that an environmenthas to offer. There is an approach called ARGO that allowsagents to collect data directly from sensors and process themas beliefs in their Belief-Desire-Intention (BDI) reasoningcycle [9]. This process requires reading all the sensors duringevery cycle execution even if the data are not necessaryfor the agent, at that moment. Some filtering techniques areavailable, but they can only be applied after the data has beencollected [10].

Initial laboratory experiments for BDI agents in a Web-of-Cell context [11] and a proposed model of many resourcesof the factory following the A&A [1] are works that considerphysical environments using the notion of artifacts. However,both implementations are domain-specific. Artifact is a suit-able notion for agents to interact with physical objects in aCPS. When MAS employs artifacts, agents are able to accessthe physical environment according to their need. It avoidsthe agents to collect unnecessary data. However, traditionalagent-oriented programming languages do not provide directapproaches to access physical environment and they are lim-ited to a particular application domain.

Some initiatives, like the Predictive Maintenance Program(PMP) reveal the importance of collecting data from sensorsin the environment to perform predictive maintenance [12].This importance can also be seen in the Oil & Gas industrybecause predictive maintenance can minimize economic andenvironmental losses from poor preventive maintenance.

The objective of this work is to provide physical artifactsfor interfacing hardware devices from a physical environmentto be accessed by MAS in a CPS. In order to developtheses Physical Artifacts, it will be created an extension of

DOI reference number: 10.18293/SEKE2020-154

CArtAgO artifacts that communicates with microcontrollersusing serial interfaces. A case study will be presented in ascenario considering a physical engine as an artifact in theOil & Gas field.

This paper is organized as follows: Section 2 presents thetheoretical background to understand the idea; Section 3 showsthe methodology used to implement Physical Artifacts, Cen-tralized Layer, and the scenario of study; Section 4 presentsthe related works and the Section 5 concludes this work.

II. THEORETICAL BACKGROUND

Multi-Agent technologies provide tools for distributed con-trol, decentralization, adaptation, and openness. These char-acteristics can be found in four MAS domains: (i) agent-oriented programming languages, (ii) interaction languagesand protocols, (iii) environment frameworks, architectures andinfrastructure, and (iv) organizational systems. These perspec-tives lead MAS to the four dimensions of development, suchas described by JaCaMo approach [13] and complementaryworks [6]: organization, where rules and missions are definedto ensure the society behavior; agent, where BDI agentsare implemented; and environmental, responsible to integratethe external environment and agents using artifacts withoperational functions; integration, that represents programlanguages responsible to ensure integration between agents,artifacts and organization rules.

In the Multi-Agent field, artifacts are Activity Theory andDistributed Cognition-based computational devices existing inenvironments and capable of performing a particular functionor service that agents can explore. Regarding the agent/artifactrelationship, there are two different types of external objectivesattached to an artifact: (i) use value, where external goalshead the artifact selection by agents; and (ii) use, which isassociated with agent’s internal goals [14]. Therefore, threedistinct aspects characterize the relationship between agentsand artifacts: agents can select, use, and construct/manipulateartifacts, where the latter occurs when the artifact does notexist and needs to be created.

Artifacts are composed of four elements [13]: User Interface(UI), Operating Instructions (OI), Function, and Structureand Behavior. User Interface (UI) is a set of operations thatagents can call to use the artifact; Operating Instructions (OI)describe how the artifact should be used to access its func-tionality; Function is the purpose of the artifact’s existence;and Structure and Behavior are the internal characteristics ofartifacts that define how it is implemented [15].

For programming the environmental dimension for agents,there is the CArtAgO framework, which is based on threemain pillars. The (i) Agent Body is the part of an agent whereartifacts represent some behaviors that it can access and controlbut it is not part of their internal reasoning; (ii) Artifactsare the components identified in a Workspace that agents orany part of their body can interact with; A (iii) Workspaceis used to define the desktop topology. Artifacts and AgentBodies are stored in these Workspaces, where the relationbetween them is established. Then, artifacts must be within

a specific Workspace so that agents can use. Consequently,events generated by these artifacts can only be seen by agentsliving in the same Workspace [16].

Using artifacts that are only accessible within theirworkspaces may not represent the best approach to be em-ployed in dynamic scenarios since it restricts agents that arenot originally from these workspaces to access the environ-ment’s resources. In dynamic scenarios agents can come andgo freely and they can compete for each available component.Moreover, the environment should be open for any entitythat intends to enter it. However, even CArtAgO, and otherlanguages and frameworks that consider the development ofartifacts do not provide a distributed and open characteristicsfor environments.

III. METHODOLOGY

In CPS, the use of environmental objects by computationalentities is an essential factor that helps these entities toadapt to environments with dynamic characteristics. Besides,these environments are increasingly demanding automation,pro-activity, and cognition. While the agents layer promotescomputational intelligence and the Organization layer pro-motes social rules, the Artifacts layer encourages the mod-eling of objects from the external environment. Although theenvironment layer is ideal for representing objects from theexternal MAS environment, there are approaches that stilltransfer this responsibility to agents. Therefore, this workpresents a solution to apply MAS in physical, dynamic, andintelligent environments using Physical Artifacts to connectMAS artifacts to ATMEGA microcontrollers. A scenario willbe presented with instrumented engines in the Oil & Gasindustry with a focus on predictive maintenance implementedin MAS with Physical Artifacts.

A. Oil & Gas Engine Scenario

When it comes to equipment maintenance, the naturalapproach is prevention, which aims to replace defective com-ponents or parts from time to time. However, this type ofmaintenance can be costly from a financial and environmentalpoint of view. From an economic point of view, the periodicreplacement of a specific component can make the processmore expensive; from the environmental point of view, theequipment may present failure situations before the replace-ment period and cause accidents to the environment. On thefinancial side, prediction is better than prevention becausepredicting that the equipment’s life cycle will be longer thanusual can avoid spending on unnecessary maintenance. Onthe environmental side, predicting that equipment is beingdamaged can result in support before it is damaged. Followingthis idea, the Oil & Gas industry benefits in the economic andenvironmental fields with predictive maintenance.

The Strategic Petroleum Reserve (SPR) - that is an Oilemergency fuel storage unit - is composed of several enginesthat supply power to the pumps that move a large amountof oil in the unit. As the SPR does not have a continuousoperation, the motors do not remain connected at all times,

which hinders the temporal precision that is necessary to carryout preventive maintenance. Therefore, prediction techniqueson engines such as vibration analysis, thermography, and oilanalysis can be useful to reduce maintenance costs and preventaccidents. In addition to sensors for analysis, the motors haveactuators that define their operation and can also be activatedintelligently to minimize the risk of equipment degradation.

As a motivation to use prediction as an approach, thePredictive Maintenance Program (PMP) proposed in 1994sets targets for reducing maintenance costs by 20% by thethird year of operation of this PMP [12]. With PMP, itis possible to offer accuracy to equipment operators as towhen intervention should occur. In this case, expenses withunnecessary maintenance and the risk of accidents would bereduced.

B. The Physical Artifacts

A Physical Artifact is an extension of the standard MASArtifact capable of integrating with a physical Device inthe environment to collect its sensor data or send actuationcommands to actuators. For Physical Artifacts, a Device is anobject in the physical environment composed of a microcon-troller with sensors or actuators. Besides, a Device must havecommunication functions between the microcontroller andanother external computational entity to provide readings on itssensors and receive commands for its actuators. Therefore, tobecome a Physical Artifact, an object in the physical environ-ment must assume the characteristics of a Device. In this case,the Operation Functions of this Artifact can be implementedto read the sensors and operate directly on the actuators ofthis Device. For example, in the scenario of engines in the Oil& Gas industry, it is necessary a microcontroller in them thatsends the data from the vibration sensors, thermography andoil to this Artifact.

To create Physical Artifacts, the Artifacts implementationof CArtAgO framework was employed. We chose CArtAgObecause it is used to create the environmental dimension ofthe JaCaMo framework for Jason and because both CArtAgOand Jason are widely used in the academia. In the hierarchicalstructure of CArtAgO, the Physical Artifact is a child class ofthe Artifact class. Therefore, physical artifacts can also imple-ment Observable Properties and Operations. It is expected withthis integration to allow MAS integration with CPS withoutoverloading agents.

Once incorporated as CArtAgO Artifacts, Physical Artifactsmust be able to communicate with Devices in the physicalenvironment. For this, the serial interface Javino [17] wasemployed, which is a library that implements a protocolfor exchanging messages between low-level hardware (mi-crocontrollers) and high-level software (Java). The choice forJavino is justified because it is a serial communication librarythat handles error detection, unlike libraries based on serialports, such as RxTx and JavaComm. The messages exchangedbetween hardware and software follow a format composed of3 fields: 2 bytes of a pre-scope that is used to identify thebeginning of the message, 1 byte to represent the size of the

main content of the message, and finally, 256 bytes containingthe content of the message to be passed. The loss or collisionof information from the past message is verified through thepre-scope field and the size field: the receiver validates thecontent in the pre-scope; if the preamble is correct, the sizefield helps to verify that the message arrived at the correctsize. If all verification is validated, the message is used;otherwise, the message is discarded. Javino offers three modesof operation: Send, Request, and Listen modes. The Sendmode provides simplex message transmission from softwareto hardware; the Request mode offers half-duplex messagetransmission (the hardware responds to the message sent); theListen mode allows the transmission of simplex messagesfrom hardware to software. Another factor that justifies thechoice of Javino is the possibility that it is designed to bemulti-platform and can be used in ATMEGA, PIC, or IntelFamilies microcontrollers.

The use of Javino as a connection bridge must be analyzedboth on the side of the abstraction (Physical Artifact) andthe embedded hardware (Device). On the Physical Artifactside, the Javino implementation class for high-level softwareis added as an attribute to the PhysicalArtifact class andinstantiated directly in the constructor. All child classes ofPhysicalArtifact must define, via abstract method, the fol-lowing values: Serial Port that will be used to connect theartifact to the microcontroller (method String definePort()),a Number of Attempts to send a message (method intdefineAttemptsAfterFailure()), and Timeout in millisecondsbetween one attempt to send a message and another intdefineWaitTimeout()). In addition, the class PhysicalArtifacthas the implementation of the String read() method, whichperforms reading from a physical device in Javino Listenmode; and also has implementation of the void send (Stringmessage) method, which sends messages in Send mode to themicrocontroller. Figure 1 shows the architecture of PhysicalArtifacts in a MAS communicating with a physical environ-ment.

Fig. 1. The architecture of a MAS integrated with a physical environmentshowing only the Agent level and the Environment level. At the Environmentlevel, Standard Artifacts are together to Physical Artifacts that connects to aDevice in the physical environment using Javino middleware.

C. Engine Scenario Prototype

To represent the engine scenario in the Oil & Gas industryand test the Physical Artifacts approach, a prototype of aninstrumented engine with a sensor was created and connectedto a MAS that will control it, as shown in the Figure 3. Thephysical prototype consists of a fan to represent the motoractuator, a temperature sensor, and LEDs that indicate the stateof operation of the motor. An Arduino Mega was used asa microcontroller that contains all the sensors and actuatorsof the prototype. Besides, Arduino Mega is responsible forexchanging messages with MAS. This physical configurationconfigures the physical prototype as a Device that can be usedby a Physical Artifact.

The engine designed in the prototype has the followingoperations: turn on, off, block use, unlock use. In particular,the blocking operation is used by operators when the engineis in an abnormal condition and should not be operated. Inthis prototype, the motor has three possible states: Readyto be Operated, represented in the first lower frame of theFigure 3, where the prototype is turned off and unlocked;On, represented in the second lower frame of the Figure 3(represented by the connection of two of the three LEDs);Blocked, represented in the third lower frame of the Figure 3,where the motor is blocked for use (indicated by the red led).

On the Arduino side, the Javino library is imported andused as a support in sending and receiving messages to thePhysical Artifact. The Arduino was programmed to send datafrom the temperature sensor whenever a message arrives fromthe Artifact that requests it. In addition, the Arduino operatesthe engine whenever the Physical Artifact requests one of theavailable forms of operation.

On the MAS side, the Motor Artifact described in Figure 2was created that extend a Physical Artifact. In this MotorArtifact, operations are implemented to read the temperaturesensor, turn on, off, lock, and unlock the motor. In addition,an Agent Manager was created in MAS to control operations.For this, this Agent creates the Artifact Motor and starts abasic cycle of activities to test all the operations provided byArtifact. In the upper left corner of Figure 3, the running agentlog is displayed.

When modeling the class diagram in Figure 2, a repre-sentation of the engine for the system with the respectiveregistration information can be seen. Besides, a model ofsensors and sensor measurements was created to record thedata in a MySQL database. With this, an application wasdeveloped to allow monitoring at the level of the engineoperator so that it can visually diagnose the engine situation.In the upper right corner of the Figure 4, is showed a graphwith temperature measurements of the environment where theprototype is located. These measures is in degrees Celsius unitand are provided by the Physical Artifact that reads the engine.The variation in the graph can be analyzed by an operator inthe field of work and serve as a variable in the generation of aprediction diagnosis, for example, associating that the increasein temperature crossed with other data, means loss of engine

life.

Fig. 2. Class diagram of the model made to represent the Engine scenario.

Fig. 3. Engine scenario in the Oil & Gas industry in execution: AgentManager performing control and monitoring of the Physical Artifact, andprototype of the engine connected to the MAS.

D. Experimental Evaluation

From the elaborated scenario, tests were done to validate thefunctioning of Physical Artifacts in the physical environment.For this, the requirements of the framework were raised tosupport the experiments: (i) the Physical Artifact must be able

Fig. 4. An application that displays a line graph with temperature measure-ments of the environment in which the prototype is based on time, in degreesCelsius, provided by the Physical Artifact. The X-axis is expressed in hours,minutes, and seconds. The temperature measurement is an example, whichcould be replaced by measurements from other sensors.

TABLE ITHE CASE STUDY DESIGN

Design DescriptionObjective Analyze the functioning of the MAS Physical Artifact in

a physical setting.Case The Physical Artifact will be connected to an oil and gas

engine that must be monitored and controlled to performpredictive maintenance.

Questions Is the Physical Artifact capable of sending and receiv-ing information through Javino? Do physical Artifactsrespond to agents’ requests in up to one second? DoPhysical Artifacts stay running for a minimum of 24hours?

Method Observation method with a low degree of researcherinteraction.

to send and receive information using Javino; (ii) the PhysicalArtifact must be able to respond to agents’ requests within onesecond, which is considered acceptable within the high-levelprogramming field; (iii) the Physical Artifact must be able tofunction in a 24-hour period in the worst case.

Based on these premises, routine tests were carried out,where the agent requested the operations to start, stop, restart,lock, and unlock the engine. At each operation, the agentrequests data from sensors ten times. It was concluded thatthe commands from the Artifact work normally. In addition, athroughput test was carried out between the agent’s commandand the execution of the Artifact, and it was observed that thewaiting time is below one second. Finally, the Agent Managerwas kept in operation for 24 hours, where it was observedthat the Artifact continues to respond with a failure rate of0%. Table I shows the case study analysis of this scenario andTable II shows the results from tests.

IV. RELATED WORKS

Physical environments have been demanding computationalsystems more proactive, autonomous, and adaptable to solve

TABLE IIEXPERIMENTAL EVALUATION RESULTS

Test Description ResultThe connection betweenPhysical Artifact andMicrocontroller

Percentage of success (%)when exchanging data with themicrocontroller

100%

Physical Artifact Re-sponses to the Agent

Maximum time (milliseconds)that an Agent takes to receivedata from the Artifact

1000ms

Physical Artifact execu-tion time

Checks whether the PhysicalArtifact remains running for24 hours

Yes

increasingly complex problems. The community has beendeveloping some works using MAS in industry as an attemptto increase pro-activity and autonomy in the production chain.There is a work in the Oil & Gas industry which usesBDI agents to filter alarms that are generated by differentconditions [8]. This filtering considers that an operator isnot able to observe a broad set of alarms and act on them.Besides, excessive alarms can hide an important occurrence,and therefore there must be an intelligent system capable offiltering this data. For this, an alarm management system wasdeveloped using agents able of reading sensors and act ondevices. However, agents were programmed directly connectedto environments — in case of agents responsible for only onesensor or only one actuator — without using the notion ofartifacts. As a result, agents could face bottlenecks in theirreasoning due to the need to be continuously collecting datawithout necessarily using it.

ARGO [9] is a customized Jason agent’s architecture thatallows interactions with physical devices such as sensors andactuators. For this, a serial interface between microcontrollersand Java programming language was developed to collect alldata from the environment to be added to the agent’s beliefbase. The generated data flow overloads ARGO agents andfiltering techniques [10] can be employed to select whichperceptions the agent has to focus on. However, the sensorsand actuators are available only for a specific MAS and theyare not shareable. Besides, ARGO may experience a decreasein computational performance as the amount of informationto be perceived increases. Both works could benefit froman approach that exposes sensors and actuators as shareableresources in the IoT.

Given the overload on the agents and aiming to take ad-vantage of the MAS environmental modeling resources, someworks developed solutions applied to physical environments.In the energy sector, a Web-of-Cell (WoC) approach [11] usesMAS to help design and test distributed solutions. For this, theJason framework is used to develop BDI agents; environmentmodeling is done using the CArtAgO framework, which allowscreating a bridge between the agent layer and the environmentlayer. The communication between the modeled environmentand the physical environment was done by the communica-tion infrastructure of the intelligent and configurable networklaboratory (SYSLAB). However, this communication bridge is

strongly linked to the SYSLAB structure, which still does nothelp in the mission to facilitate implementations with MASthat involve environmental modeling.

MAS heterogeneity has been increasingly required in chal-lenges that integrate CPS with environmental resources. Inthe Industry 4.0 concept, equipment and sensors must beintegrated into the same system using the most diverse com-munication protocols. Camel Artifact [1] is a componentthat uses Java-based message routing and mediation tech-nology (Apache Camel) in artifacts. A CamelArtifact makesit possible to transform physical devices into Artifacts in amore generic way than the WoC approach because severalcommunication protocols can be used to create the bridgebetween the physical and the computational environment. Forthis, routing is done that directs the messages from a deviceto the specific artifact. However, although this work does notdepend on a particular protocol of communication betweenphysical devices and MAS, there is still a strong dependenceon Apache Camel technology that guarantees message routing.Perhaps, an approach that integrates artifacts with microcon-trollers can offer even more heterogeneity because it will allowconfigurations of these devices more directly and at a lowlevel. If these artifacts were shareable between different MAS,the collected data would become resources of the environmentthat agents from any MAS could exploit.

V. FINAL CONSIDERATIONS

Normally, MAS applications using physical environmentsfor CPS overloads agents with data coming from sensors andactuators. Besides, when they are not overloaded, the connec-tions to these kind of artefacts are bounded to the providedsolution. Based on that, this work presented an extension ofCArtAgO for providing Physical Artifacts without generatingoverload to agents using a serial interface for communicatingwith heterogeneous microcontrollers.

In order to create Physical Artifacts, several technologieswere employed such as Jason and CArtAgO frameworks,the Serial Interface Javino, and microcontrollers. CArtAgO’sArtifact was extended to allow Javino to interact with sensorsand actuators connected to microcontrollers. The proposedextension was tested in an engine scenario for Oil & Gasdomain. The results showed that our approach is suitable fordesigning CPS using MAS and Physical Artifacts.

A future issue to be considered is that the fact that artifactsto be accessible only within their workspaces can make itchallenging to implement in dynamic scenarios because thisrestricts agents that are not from this MAS. If artifacts couldbe accessed by agents from another MAS, it could be possibleto create a multi-purpose layer of physical artifacts to beconsumed by different agents. As future works, we intend tocreate a shareable layer of artifacts to be used along with theInternet of Things. Agents from different MAS, or any othertechnology, could compete for Physical Artifacts. Besides, wewill extend the scenario of motors for Oil & Gas to allow amiddle layer capable of managing plans and rules for somesituations when using those motors.

REFERENCES

[1] C. Amaral, S. Cranefield, J. Hubner, and M. Roloff, “Giving camelto artifacts for industry 4.0 integration challenges,” Lecture Notesin Computer Science (including subseries Lecture Notes in ArtificialIntelligence and Lecture Notes in Bioinformatics), vol. 11523 LNAI,2019.

[2] T. Sanislav, G. Mois, S. Folea, L. Miclea, G. Gambardella, andP. Prinetto, “A cloud-based cyber-physical system for environmentalmonitoring,” in 2014 3rd Mediterranean Conference on EmbeddedComputing (MECO), pp. 6–9, IEEE, 2014.

[3] S. Wang, J. Wan, D. Zhang, D. Li, and C. Zhang, “Towards smart factoryfor industry 4.0: a self-organized multi-agent system with big data basedfeedback and coordination,” Computer Networks, vol. 101, pp. 158–168,2016.

[4] B. Vogel-Heuser, C. Diedrich, D. Pantforder, and P. Gohner, “Couplingheterogeneous production systems by a multi-agent based cyber-physicalproduction system,” in 2014 12th IEEE International Conference onIndustrial Informatics (INDIN), pp. 713–719, IEEE, 2014.

[5] O. Boissier, R. H. Bordini, J. F. Hubner, A. Ricci, and A. Santi,“Multi-agent oriented programming with jacamo,” Science of ComputerProgramming, vol. 78, no. 6, pp. 747–761, 2013.

[6] M. R. Zatelli and J. F. Hubner, “The interaction as an integrationcomponent for the jacamo platform,” in International Workshop onEngineering Multi-Agent Systems, pp. 431–450, Springer, 2014.

[7] E. A. Lee, “Cyber physical systems: Design challenges,” in 2008 11thIEEE International Symposium on Object and Component-OrientedReal-Time Distributed Computing (ISORC), pp. 363–369, IEEE, 2008.

[8] N. Sanchez-Pi, L. Leme, and A. Garcia, “Intelligent agents for alarmmanagement in petroleum ambient,” Journal of Intelligent and FuzzySystems, vol. 28, no. 1, pp. 43–53, 2015.

[9] C. E. Pantoja, M. F. Stabile Jr, N. M. Lazarin, and J. S. Sichman,“Argo: A customized jason architecture for programming embeddedrobotic agents,” Fourth International Workshop on Engineering Multi-Agent Systems (EMAS 2016), 2016.

[10] M. F. S. Jr, C. E. Pantoja, and J. S. Sichman, “Experimental analysis ofthe effect of filtering perceptions in bdi agents,” International Journalof Agent-Oriented Software Engineering, vol. 6, no. 3-4, pp. 329–368,2018.

[11] D. Issicaba, M. Rosa, A. Prostejovsky, and H. Bindner, “Experimentalvalidation of BDI agents for distributed control of electric power grids,”in 2017 IEEE PES Innovative Smart Grid Technologies ConferenceEurope, ISGT-Europe 2017 - Proceedings, vol. 2018-January, pp. 1–6,2018.

[12] R. J. Murry and B. F. Mitchell, “Cost savings from a practical predictive-maintenance program,” in Proceedings of Annual Reliability and Main-tainability Symposium (RAMS), pp. 206–209, IEEE, 1994.

[13] A. Ricci, M. Viroli, and A. Omicini, “Programming mas with arti-facts,” in International Workshop on Programming Multi-Agent Systems,pp. 206–221, Springer, 2005.

[14] R. Conte, C. Castelfranchi, et al., Cognitive and social action. GarlandScience, 2016.

[15] A. Omicini, A. Ricci, and M. Viroli, “Coordination artifacts as first-classabstractions for mas engineering: State of the research,” in InternationalWorkshop on Software Engineering for Large-Scale Multi-agent Systems,pp. 71–90, Springer, 2005.

[16] A. Ricci, M. Viroli, and A. Omicini, “Cartago: A framework for proto-typing artifact-based environments in mas,” in International Workshopon Environments for Multi-Agent Systems, pp. 67–86, Springer, 2006.

[17] N. M. Lazarin and C. E. Pantoja, “A robotic-agent platform for em-bedding software agents using raspberry pi and arduino boards,” in 9th

Software Agents, Environments and Applications School, 2015.